Opportunity Identification

What Opportunity Identification is

Opportunity identification is the third step of an AI audit. By this point the processes have been mapped and written down. This step asks of each one what AI can do reliably against it and, as importantly, what it cannot. What comes out is a prioritised roadmap, with each opportunity carrying the reason it sits where it sits.

It is the step that turns a description of how work runs into a decision about what to do first. The AI audit page covers the whole sequence, and this page covers the one step where the ordering is made.

Where it sits in the Audit

An audit runs five steps in order: map your processes, document how they run, identify the opportunities, test a prototype, and build and scale. Each step produces the thing the next one needs. You cannot rank opportunities you have not documented, and you cannot document a process nobody has mapped.

That is why the third step is not a brainstorm. It works on documented processes, in the organisation’s own vocabulary, with their exceptions and volumes written down. It is also the point where the audit stops describing and starts recommending.

What goes in

The input is a set of candidates: the documented processes, or parts of them, that might be worth changing. The discovery call names which processes are in scope in the first place, and on the discovery call the site says the processes worth putting inside a piece of work are named alongside the ones worth leaving alone.

By step three each candidate is a real piece of work with a sequence, decision rules and a volume, and not an idea. That is what lets the step weigh them against each other instead of guessing.

The four things each candidate is weighed on

The audit weighs each candidate on four things. The first is how often it repeats. A process that happens constantly has more to give back than one that happens twice a year. The second is what the manual version costs in effort, meaning the time and attention a person spends on it today.

The third is whether the data it needs exists in a usable state. A process that runs on scattered spreadsheets and one person’s memory has a data problem no model fixes, and the audit names that and does not recommend a tool on top of it. The fourth is what regulatory exposure attaches to it, which is covered in its own section below.

The weighting is yours, not a Template

The audits page shows the four as an illustration, with bars for volume of work, exception rate, data available and compliance load. It labels the panel illustrative and says the weighting is set against your own documented processes in step three and not from a template.

That is worth taking at its word. The four things are real, and the bar lengths in the picture are not a measurement of anything. A business running two or three customer-facing processes gets a ranking built around those, and not a generic score copied from elsewhere.

What AI can do reliably, and what it cannot

The audit does not only ask where AI could be used. It asks what AI can do reliably against each process and what it cannot. The second half of that sentence carries as much weight as the first. A candidate can be technically possible and still a poor use of effort, or a poor fit for the way the work actually behaves.

This is also where the ladder of delegation comes in. The site’s article on automating a task against automating a decision sorts work onto four rungs: Assist, Recommend, Execute and Never delegate. It says every real workflow is a chain of steps, each belonging on its own rung, and that the rungs are decided by five questions and not by what the technology can attempt.

The five questions behind the Ladder

The five questions are how often the work happens, what a wrong outcome costs, whether it can be undone, how clear the governing rules already are, and whether the outcome needs a kind of legitimacy a model cannot hold. The article compares a password reset with a candidate shortlist. The reset happens constantly, is reversible in seconds and follows an unambiguous rule, so it climbs to Execute.

The shortlist happens less often, is hard to undo once a candidate has been told no and is judged against criteria never perfectly fixed in advance, so it stops at Recommend. Two pieces of work that look alike land on different rungs for reasons that have nothing to do with which one AI could technically do. The ranking uses the same reasoning.

Set aside, with the reason written down

Candidates that are technically possible but not worth doing are named as such and set aside with the reason recorded. The site’s point is that a list that says yes to everything is not a prioritised list. An audit that recommends everything has not done this step.

Setting a candidate aside is a result, not a failure. The reason is written down so the decision can be revisited when things change, and so nobody has to argue the case again from scratch.

Lower on the list, not off it

The audit article adds a useful subtlety. A change that scores well on paper but fails the people test, meaning the team would not actually use it, sits lower on the list and not off it. It may still be worth doing once a simpler item has built confidence in the process.

So the ranking is not only about value. It also weighs effort, data and whether the people who would use the result can and will. The order is a sequence, and a good sequence puts an easier win ahead of a bigger but harder one when the easier one earns the trust that makes the harder one possible.

When the Answer is not AI

The article says a genuine audit is prepared to find, when it is true, that the process examined is not an AI problem at all. Its examples are a pricing issue, a communication gap and a sales process losing leads before any tool would ever touch them. It adds a plain line: no audit fixes that by recommending software.

That is the honest end of the step. A ranking that puts a tool against a problem it cannot fix would still look like a result, which is why the step is written to allow for the finding that AI is not the answer.

Regulation is weighed inside the ranking

The EU AI Act and GDPR obligations are assessed here, inside the ranking, and not added as a review at the end. Regulatory exposure is one of the four things that decide where an opportunity sits, so an opportunity that carries heavy obligations is ranked knowing that.

The audit also produces a compliance checklist covering the obligations attached to the opportunities ranked highest. It identifies and checks against those obligations. It does not certify them, it is not a legal opinion, and no audit substitutes for advice from a qualified lawyer. The high-risk AI system page covers the class of use that carries the most.

What your side does

The audit page says your side has a part in this step. You set the constraints that are real and not merely preferred, and it gives three examples: data that cannot leave a jurisdiction, decisions that must keep a human on them, and systems that cannot be touched this year. Then you confirm the ranking or challenge it.

The distinction between real and preferred matters. A real constraint changes the ranking. A preference can be argued with. Naming which is which early keeps the roadmap honest, and it stops a recommendation being built on something that turns out to be immovable.

What comes out

Two things. The first is a prioritised roadmap of opportunities, each carrying the reason it sits where it sits. The second is a compliance checklist covering the EU AI Act and GDPR obligations attached to the ones ranked highest.

The site’s advice on the shape is that a short ranked list is more useful to act on than a sprawling document nobody finishes reading. It also advises asking to see the ranked list before committing to a tool, and treating an audit that hands you one without being willing to say no to anything as a warning sign.

What comes after

The output is a plan and not a finished workflow. The next step builds the highest ranked opportunity as a narrow working prototype and puts it in front of the people who would use it, running on real work or a realistic sample of it. It exists to answer one question: does this behave well enough on this organisation’s own work to be trusted with it?

Building a full AI workflow from a ranked opportunity is separate work, with its own scope and guardrails. Workflow design is the stage between naming an opportunity and building anything. The agentic AI page covers why the site tests narrow before building wide.

A worked example

This is an illustration, not a real client. A small firm brings four candidates out of the earlier steps: a monthly report built by hand, client onboarding paperwork, a shared support inbox and a hiring shortlist. The report repeats every month, costs a person a day of effort, and its data sits in three exports that are usable.

The report ranks first. The onboarding paperwork ranks second, held back because its documents vary a lot. The inbox is set aside with the reason recorded: it is low volume and would not repay the effort. The shortlist is kept, but as Recommend with a named reviewer, and its compliance load is noted in the checklist because it feeds a decision about a person. The firm challenges one ranking, on the grounds that a system cannot be touched this year, and the order is adjusted. Nothing was built. The firm has a roadmap with reasons.

Mix-ups worth avoiding

Four confusions come up often enough to name. The first is treating it as brainstorming, when it works from documented processes and not from ideas. The second is treating the ranking as a list of things to build, when it is a sequence with reasons and a set of things set aside.

The third is treating a high ranking as a promise, when the prototype is what justifies a build. The fourth is treating regulation as a separate review that happens at the end. The site says it is weighed here, inside the ranking.

Where it goes wrong

The step fails when the earlier ones were skipped, because you cannot rank what you have not documented. It fails when the list says yes to everything, when regulatory exposure is left out until after the ranking is agreed, or when the people who would use a recommendation are not asked whether they would.

It also fails when readiness is ignored. The site’s article on what predicts whether an AI project reaches production says an audit is where readiness is assessed and sequenced, and names a sponsor, a success measure and an internal champion as traits to check. A high-ranked opportunity without those is a sequencing question and not a stop sign.

Where it appears on the site

Opportunity identification is named as a stage on the home page’s adoption sequence, and it is step three on the how we audit page. The what an AI audit actually checks article explains how a mapped process becomes a ranked list, and the well-scoped workflow article explains why scope decides the result.

Every audit starts with a discovery call, and the step sits inside the AI audit.